Yuriy, if you place your stitching vias (or whatever you call them), close to the signal via, they will also define the inductance and impedance of the return path. The inductance of the complete arrangment (signal via, stiching vias/return path) depends almost entirely on the geometry of the vias and the distance between the stitching vias and signal via. If you increase your simulation area without increasing the geometry/position of the stitching vias, then the inductance/impedance of your return path will not change. So your explanation that changing the simulation area automatically changes the impedance of the return path and |S11|, is not correct. If you do not believe me, then contact Dr. Howard Johnson for further explanations. In his book (advanced black magic) he has an excellent explanation of the relationship between these stitching vias and the inductance of the return current of the signal via. This is the only text I know that treats these issues in a very practical manner and I think it will be of great help to you. I strongly recommend it to every other person working with via modeling. Note that the number of stitching vias you use depends on a number of other factors. From his book, you'll see just a method. You can apply the method to your own designs. Don't expect an automatic solution to your problem! Remember, Maxwell never said Ampere is wrong. He said Ampere's circuital law is not complete and then completed it using the displacement current term (See J.C. Maxwell - On Physical Lines of Force, 1861). Now, when your stitching vias define the return current path, the conduction current is much more greater than the inter-plane displacement currents, such that we can use the original form of Ampere's law without any loss of accuracy. Quasi-static methods use this approximation too. It's not that the displacement current is not present. But it is too small to be considered. With that said, you can visualize the discontinuity (considering its return paths and traces) as consisting of 2 parts. The closest part (which you beautifully describe as the "near field zone") and the "far-field zone" - if you allow me to use your words. In the "near field zone", which should be less than 3mm for frequencies up to 50 GHz (from my experience) when using typical PCB/package geometries, |S11| (at a given constant frequency) depends almost completely on the higher order modes excited by the discontinuity. |S11| decreases as the length increases because the higher order modes die off. In the far-field zone, |S11| increases because the higher order modes have compeltely decayed and other factors now play the deciding role. As you know, |S11| for a given line length at a givn frequency is constant. This is the method I have been using to define the boundaries of both vertical discontinuties(such as via-holes) and horizontal discontinuties (such as bends). And I'm fine with it. I have done many simulations and measurements. As I said earlier, I got it from the paper I cited on Nov. 20. and other publications from these authors. I had problems to accurately model discontinuties at higher frequencies (up to 77 GHz) years ago. Remember it is just a method and you have to take into consideration your real design enviroment, if you apply the method proposed in the paper. Don't expect an automatic solution to your problem! As far I know, the term "boundaries of discontinuities" as used in PCB/package design was first introduced by these authors. If you have your own method, then you can also use it. But one thing is clear. You have to define these boundaries at your 20 GHz and beyond. If not, your models are not correct and can not be validated with measurements. I'm not interested in your tool. Just like Chris, I'm interested in methods or methodology as he calls it. Hope this helps Best regards Charles Yuriy Shlepnev <shlepnev@xxxxxxxxxxxxx> wrote: Chris, I think all definitions of the interconnect models you provided may be applicable under different circumstances. It depends on possibility to localize discontinuities such as vias and splits for the electromagnetic analysis. If all discontinuities in you channel can be isolated for the electromagnetic analysis, then the de-compositional analysis you described in a) and b) can be safely used. If such localization is impossible, the system-level model has to be built either on the base of a complete 3D electromagnetic analysis of the whole board (possible but not practical) or, alternatively, the de-compositional model has to be extended with models of such structures as parallel planes and splits with all decoupling structures connected to them (hybrid system-level models with transmission planes). Those structures are the major reasons of non-localizability of discontinuities on PCB and in packaging applications. How to define the boundaries of a discontinuity and localizability. If simulation results (S-parameters for instance) are relatively independent on the simulation area size and on the boundary conditions, the discontinuity can be isolated for the analysis and a reusable model can be generated (otherwise the discontinuity is not localizable). Simple rules based on line width/substrate height can be used to define the simulation area in case of localizable discontinuities. Phase reference planes may be shifted toward the discontinuity to make it electrically smaller (for better fitting or interpolation). A line segment with a minimal length to prohibit interaction through the higher order modes have to be added at the system-level in that case. This de-compositional technique used in microwave engineering since 40-s (Levin, Advanced theory of waveguides, 1951) and is the mainstream in the microwave system-level analysis tools. The smaller the localizable discontinuity the smaller the effective discontinuity area. It provides good models even for micron-sized structures up to sub-mm wave frequencies. Note that dependency of S-parameters on the simulation area sometime has nothing to do with the higher order modes discussed here before, but rather related to parallel planes and to localizability. S-parameters of a single via without or with a stitching via nearby can show significant dependency from the simulation area simply because of the discontinuity is not completely localized and the impedance of a cavity formed by parallel planes may change the |S11| for instance. Increasing the simulation area, one increase the inductance of the return path and together with the via capacitance to the planes it may be visible as the decrease of |S11| with the increase of the simulation area size (effect observed in the paper cited below). Such problems may be on the border line between the localizable and non-localizable problems and sometime may even require the system-level hybrid models with the transmission planes. Who has to define the boundaries of discontinuities or minimal length of the line segments to connect the discontinuities. Ideally, it has to be the system-level tool that decomposes a channel into transmission lines, discontinuities and possibly transmission planes. All mainstream SI tools are already based on the decomposition into line segments. It may include coupling between the lines bases on physical or electrical thresholds. The same approach has to be used to define what discontinuities may be analyzed with a 3D EM tool and what discontinuities require hybrid models. If two discontinuities are too close to each other (physical or electrical criteria can be used) - they have to be analyzed in a 3D solver as a whole and so on. In addition, a 3D solver has to define sufficient simulation area automatically and produce the model that is electrically as small as possible. Without such interaction between the system-level tool and a 3D solver you have to follow the recommendations provided by a 3D tool vendor and make sure that the discontinuity model is connected in the final design with sufficient line segments. I think that report on the minimal length of the line segments would be a good feature for an electromagnetic tool. Best regards, Yuriy Yuriy Shlepnev, Simberian Inc. www.simberian.com -----Original Message----- From: Chris Cheng [mailto:Chris.Cheng@xxxxxxxx] Sent: Thursday, November 29, 2007 6:18 PM To: ch_harrington@xxxxxxxxx; shlepnev@xxxxxxxxxxxxx Cc: SI LIST Subject: RE: [SI-LIST] Re: Signal crossing Split plane Charles and Yuriy, I have a philosophical question about modeling these 3D structures. It seems both of you agree that the entry ports needs to be back out to certain distant from the structure itself (most likely dimensionally compatible to the structure itself). So what is the definition of the overall system level interconnect model ? One can have the following defintions : a) interconnect model (most likely lossy trace model) with length up to the extended port location + the 3-D model of the plane cut/via transition model b) interconnect model (most likely lossy trace model) as report by the design data base + the 3-D model of the plane cut/via transition model - effect of just the extend port length of the interconnect model c) the entire interconnect structure is simulation in one gigantic 3D structure The combine last two terms of b) is what Roger Harrington used to call excess parasitics. In a PCB interconnect environment, a) and b) for all practical purpose are the same because the interconnect length >> extended port length. But for package model where the entire structures are measured in mm or mils, a) and b) may have significant differences. Should a 3D cad tool report the "excess parasitics" so that users can simply use the length report of the design database, then add in the via/plan cut section anytime he/she encounters such structure ? Or should a 3D cad tool be just modeling the true 3D structure but then has to warn users to back out the interconnect trace length to account for the extend port length (which seems to require careful consideration of the 3D structure on a case by case basis). Or, just lump the 3D structure into a gigantic 3D file together with the rest of the interconnect and pray that the simulator will converge ? -----Original Message----- From: si-list-bounce@xxxxxxxxxxxxx [mailto:si-list-bounce@xxxxxxxxxxxxx]On Behalf Of Charles Harrington Sent: Wednesday, November 21, 2007 2:57 PM To: shlepnev@xxxxxxxxxxxxx Cc: sunil_bharadwaz@xxxxxxxxx; 'SI LIST' Subject: [SI-LIST] Re: Signal crossing Split plane Yuriy, I think we really have to end the discussion. I recommend you also talk to some experts in this forum about your models. They will tell you exactly what Im trying to say and even more. I didnt even know you have your own software. But I cannot understand why you make such claims that your software can compute whatever multilayered geometry and that it also automatically defines the boundary of the discontinuities. You know this is not true. We all know this is not true. So why do you make such claims? If I ask you, with what degree of accuracy does your software compute "whatever multilayer geometry" (when compared to measurements) and how does it automatically define the boundaries of discontinuities, I know that you will be baffled. So, I dont need the answers to these questions. However, I'm glad you acknowledge the fact that you need about 1mm distance away from the via pad at such higher frequencies to get accurate results. Let us leave it there. I will not write any more. I wish you the best with your models. 25 yrs of experience is quite a lot. I respect that. But as you can see, there is still a lot out there to learn. Best regards Charles Yuriy Shlepnev wrote: Hi Charles, Thank you for the reference. I am familiar with this paper as well as with the other publications of this group from Fraunhofer Institute. First of all, our 3D full-wave solver allows to build different via-hole models. It solves whatever multilayered geometry with ports you put in there. Second, the solver automatically defines the boundary of the discontinuities. See for instance the final model for optimal via-hole on slide 8 in http://www.simberian.com/Papers/OptimalDifViaholesDesign6pPCB.pdf. The differential line segment length in that particular example is about 1 mm, that is sufficient for high-order modes to die even at 30 GHz. Though, to define the area we use technique different from one described in the paper (I hinted details earlier). Lumped ports are often used for the preliminary optimization of via-holes because of it is quick and it provides good approximation (see for instance the final model and comments in the presentation mentioned above). Essentially, it ends the discussion. If you looked through the app notes on our web page you just saw the tip of an iceberg. We put about 25 years of research to develop and validate the technology. It is well documented on our web site in Downloads/Papers and Presentations areas. And, I do not even want to start discuss the definition of ports or multimodal decomposition, because of it looks strange to me that after reading Collins you still do not understand what it means and how it applies to the multilayered circuits. Best regards, Yuriy Yuriy Shlepnev, Ph.D. President, Simberian Inc. 2326 E Denny Way, Seattle, WA 98122, USA Tel/fax +1-206-726-1098 Cell +1-206-409-2368 Skype shlepnev www.simberian.com From: Charles Harrington [mailto:ch_harrington@xxxxxxxxx] Sent: Tuesday, November 20, 2007 2:46 PM To: shlepnev@xxxxxxxxxxxxx; scott@xxxxxxxxxxxxx Cc: sunil_bharadwaz@xxxxxxxxx; 'SI LIST' Subject: RE: [SI-LIST] Re: Signal crossing Split plane Yuriy, I agree with some of your views. However, they contradict your via models. I couldn't reply yesterday, because I was trying search for the reference I mentioned, since you needed it. Many other people replied off-line and so needed the reference. Got it from IEEE Xplore. A Novel Methodology for Defining the Boundaries of Geometrical Discontinuities in Electronic Packages Ndip, I.; Reichl, H.; Guttowski, S.; Research in Microelectronics and Electronics 2006, Ph. D. 12- 15 June 2006 Page(s):193 - 196 You mentioned in your mail that the near field zone as a result of the higher-order modes excited at the via expands with frequency and is very small. I agree with you. But the question is this. How small is it? How small or big is at 1 GHz, 10 GHz, 20 GHz? Have you ever studied it? You have to take this zone into consideration when studying vias or any other structures that excite higher order modes. The method proposed in this paper is quite illustrative and useful. I understand it this way (Please correct me if I understand it wrongly): These higher-order modes (e.g., TE, TM...) are characteristics of the trace or transmission line and they die exponentially away from the point of excitation, i.e., the via-trace interface. S-parameters, like other network parameters, give us the relation between input and output signals. Now, to obtain S11, for example, you need to get the ratio of the reflected and input signals. Both signals must be of the same "type". We can not directly compare cars and aeroplanes, though both are used for transportation. You know your input signal (e.g., a transverse electromagnetic wave), because you excited it at the port. At discontinuities, an infinite order of given higher-order modes can be excited. The orders or strength of the excited modes differ from one discontinuity to another, although the modes can be the same. So, there is no way you can know all the orders of the higher-order modes excited and how they interact. Now if you place your ports quite close to the point of excitation of these modes, then your S-parameters must be wrong. Why? In this case, to obtain S11, you need to obtain the ratio of the unknown higher-order modes and your known excited transverse electromagnetic wave at the port. That's why in most 3D full-wave solvers, it is recommended that ports should be placed far away from the discontinuities, so as to enable these higher-order modes to die. When they die, then you can easily define your S-parameters which will then be the ratio of the input signal you know (transverse electromagnetic wave) and the reflected signal you know (transverse electromagnetic wave). To define the points where these modes die or have attenuated substantially, these authors argued that near the discontinuity, the imaginary part of the Poynting vector describes the reactive energy associated with these higher-order modes. So they studied this imaginary part and used it to define the point where the modes die. I think they mentioned that only at a distance of about 1mm away from the via-trace interface, at 20 GHz (or may be 30 GHz) may you place your ports, to get correct results. Certainly, this depends on the via geometry and trace type. But I find the results very helpful and can be used as a base for further experiments. You can get the details from the paper. Unfortunately in your case, you compare what you don't know (reflected signal) and what you know (excited input signal). In your via models, neither did you define the required distance away from the via-trace interface needed for these modes to die nor did you follow the advice given in full-wave solvers to be far way from the via-trace interface. You considered the via just as the barrel and the pads at 20 GHz and beyond. That's why I mentioned yesterday that your via models are not correct and your S-parameter results are misleading. If you wish to study only the behaivor of the barrel alone at lower frequencies (for what ever reason - but not for realistic designs), then you don't even need a field solver. You can get formulas from good SI texts like that of Horward Johnson or from papers. At first I was also making the same mistakes as you are making right now. I had a lot of difficulties to correlate my simulation and measurement results. So I learnt a lot from this paper, from Professor C. Balanis (Advanced engineering electromagnetics) and from Professor R. Collins (Field theory of guided waves). I think these references will be good for you. You need all three of them. There are also a lot of points that you need to modify in your models. It's ridiculous when you talk of -30 dB attenuation of higher-order modes. Which higher-order mode? Which order of this mode? Basic electromagnetic theory teaches us that an infinite order of a given higher-order mode can be excited at any discontinuity. An interaction between makes matters worst. So how do you separate the different orders of the modes and tell which one attenuates by -30 dB? Are the modes propagating or evanescent? Never use rule of thumbs that have no base. I supposed you meant attenuation of the fundamental mode which is propagating. I don't know anything about the lumped ports you use. All I know is that some lumped ports in some field solvers assume perfect H boundary conditions on the sides. Consequently, depending you may not even capture stray fields. So you can even get the worst results with lumped ports. You can only shift your reference S-parameters plane and get accurate results if your model captured all the necessary field behavior. But you can not simulate the via and traces differently and then do some post-processing or circuit modeling afterwards and expect to get correct results at higher frequencies. The traces too are part of the "via effect" at least, at the frequencies you are interested in (20 GHz and beyond), because the stored higher-order modes give rise to additional inductances and capacitances. These inductances and capacitances can not be captured if you analyze the vias separately from their traces. Finally, the theory of multi-modal decomposition means different things to different electrical engineers. So I don't know what you mean. If you mean that different parts of a system can be analyzed separately and then put together, then it's true that it has been done for decades now. But the question is this. How do you bring the different parts together in the case where there are discontinuities like vias? How do you define the via? How small or big is your near field zone? I bet you, we have not yet understood this type of decomposition and it has not been done, or at least published for decades. Whenever we have to deal with vias and other discontinuities at higher frequencies, straight-forward modeling can not be used. Please Yuryi, don't get me wrong. I'm not trying to highlight on your errors. I have mine too, like any body else. No one is perfect. I'm just trying to raise the point that we need to be careful when modeling vias at your frequencies. I agree with most of the points you made, but disagree on the ones stated above. We learn from each other when we exchange ideas about such fundamental issues that affect our modeling results. I think that is one of the reasons why Ray and his team set up this forum. Best regards. Charles Yuriy Shlepnev wrote: Charles, I am sorry that the simulation examples were not helpful to you. I will appreciate if you send me the reference you mentioned - I am preparing to be shocked:) You are absolutely right, the via-holes are not just pads and barrels and there is no one solution that covers all possible cases. Analysis of different vias has to be done in different ways. Transition to the traces have to be almost always included in the final model for analysis of multi-gigabit channels. Moreover sometime the via-hole problem cannot be solved locally and require analysis of parallel plane structures with all decoupling structures attached (see technical presentation #1 at http://www.simberian.com/Presentations.php for more details on different structures). Considering the ports and excitation. Analysis of via-holes with lumped ports provides just rough idea about the via-hole behavior. It is similar to what you would see from a differential probe attached to the pads of the via-holes. Transition to traces and transmission line or wave-ports have to be used for the final extraction of S-parameters for the system-level analysis (I am sorry that you missed this part in app notes). Note that it is possible only for the localizable via-holes or via-holes not coupled to parallel planes in general. Such t-line ports have to be positioned at a distance from the via-hole that guaranties that the high-order modes are attenuated substantially (for practical applications we usually use -30 dB threshold at the highest frequency of interest). After such analysis, the phase reference planes of S-parameters can be safely shifted closer to the via-hole at the position where t-lines are still continuous to preserve causality (to the edges of anti-pads for instance). Such transformation does not affect the near field or high order modes around the via-holes and the final model can be safely connected with the transmission line segments in a system-level solver. Though, the model have to be used with transmission line segments with length not less than in the electromagnetic analysis (to avoid the near-field interaction between the vias and possible discontinuities). This technique called the multi-modal de-compositional analysis and used in microwave engineering for decades at frequencies even higher than 20 GHz. Note, that in typical PCB trace the cut-off frequencies for high-order modes are extremely high. 10 mil trace on 10 mil dielectric with dielectric constant 4.2 have cut-off frequency about 120 GHz, and the cross-over with the surface TM mode may happen only at 200 GHz. Before 120 GHz the high-order modes are evanescent and essentially form the via-hole near field. This near-field zone is expanding with the frequency, but at 20 GHz the area is still relatively small. Thus S-parameters only for the dominant modes can be safely extracted and used as the via-hole model. Cases when via-hole excite the non-evanescent parallel-plane modes and planes are not stitched close to the via-hole cannot be solved locally === message truncated === --------------------------------- Be a better friend, newshound, and know-it-all with Yahoo! Mobile. Try it now. ------------------------------------------------------------------ To unsubscribe from si-list: si-list-request@xxxxxxxxxxxxx with 'unsubscribe' in the Subject field or to administer your membership from a web page, go to: //www.freelists.org/webpage/si-list For help: si-list-request@xxxxxxxxxxxxx with 'help' in the Subject field List technical documents are available at: http://www.si-list.net List archives are viewable at: //www.freelists.org/archives/si-list or at our remote archives: http://groups.yahoo.com/group/si-list/messages Old (prior to June 6, 2001) list archives are viewable at: http://www.qsl.net/wb6tpu